Google's Gemini 2.5 Pro creates games and analyzes scenarios in real time

Google has announced Gemini 2.5 Pro — a model that not only analyzes data but also "thinks," bridging the gap between human and machine. Gemini 2.5 Pro can handle up to 1 million tokens of context — the equivalent of 700,000 words or 15 hours of audio recordings. However, the key innovation lies in its built-in Reasoning module, which allows the system to draw logical conclusions, solve multi-step problems, and even predict scenarios. For example, the AI can instantly generate video games from a text description: in a demo, the model created a prototype of an endless runner, adapting the mechanics to user requirements.

In the Humanity's Last Exam, which tests knowledge and analytical abilities at the edge of human capacity, Gemini 2.5 Pro secured first place with a score of 18.8%, outperforming previous versions. Experts note that this level of efficiency paves the way for autonomous AI agents capable of managing projects without human intervention. Like its predecessors, Gemini 2.5 Pro works with text, images, video, audio, and code, but now integrates them into a unified analytical process. This enables, for example, software optimization based on interface screenshots or subtitle generation for videos with emotional context in mind.

Test access to the model is already available to developers through Google AI Studio and to users with a Gemini Advanced subscription. The company has yet to disclose API pricing, but analysts suggest the integration will be geared toward the corporate sector.